Summary 531 & in test was sole BRAF mutation analysis. None of the reviewed diagnostics had both near-perfect sensitivity and specificity. Hence, we suggested that a multimodality stepwise approach may offer the most accurate diagnosis, by sequentially applying one sensitive rule-out test and one specific rule-in test. Limited cost-utility data were available for most of the tests, none for a multimodality approach. We concluded that geographical variations in cytology (e.g., primarily the rate of cytology suspicious for an oncocytic follicular neoplasm) and tumor genetics likely strongly influence local test performance and clinical utility, and that multidisciplinary collaboration and local implementation studies are key to decide which of the available diagnostics should locally be favored. In the supplementary data of Chapter 2, our unpublished systematic review and meta-analysis data were presented for the first time. These data were previously lost in the editorial process concerning the publication of the review included in Chapter 2. The extensive meta-analyses support the conclusions of Chapter 2, with an estimated pooled test sensitivity of 97.3% (95% CI, 94.2%-98.8%) for the Afirma® GEC and 94.2% (81.1%-98.4%) for [18F]FDG-PET(/CT), and an estimated pooled test specificity of 99.9% (99.3%-100%) for BRAF mutation analysis. Additionally, the importance of appropriate study methodology and reference standards for clinical validation studies was emphasized by critically reviewing all included studies and presenting meta-analyses for a hypothetical best-case (i.e., what if all index test results for cases with a missing reference standard were true, true-positive or true-negative) and worst-case scenario (i.e., what if all index test results for cases with a missing reference standard were false, false-positive or false-negative) for each of the investigated diagnostics. High rates of missing, appropriate reference standards (defined as histopathology, in absence of long-term follow-up) especially concerned MD studies. The best- and worst-case scenarios showed that particularly test sensitivity may vary for the worse as compared to published data (i.e., because the cases without a reference standard were oftentimes the test-negative cases). For example, the estimated pooled sensitivity for the Afirma® GEC could be 98.2% (95.5%-99.3%) in the best-case scenario but potentially only 42.5% (27.9%-58.4%) in the worst-case scenario. For [18F]FDG-PET, best- and worst-case scenarios were similar to the primary meta-analysis, as histopathology was available in all cases from all studies. Chapter 3 zoomed in on imaging biomarkers and non-invasive diagnostic imaging techniques for indeterminate thyroid nodules. Besides reviewing conventional and molecular imaging techniques, artificial-intelligence-based medical imaging and radiomic methods were discussed. These quantitative techniques appear promising by providing an objective assessment in addition to visual image interpretation. Artificial-intelligence-based imaging is currently developed on a large scale for the quantitative assessment of imaging of many different conditions. Yet, it still has a limited role in clinical practice as external validation studies are lacking. Radiomics has been investigated in indeterminate thyroid nodules in two studies, including Chapter 5 of this thesis. Both studies concluded that solely radiomic analysis on [18F]FDG PET/CT images seemed of no added diagnostic value in the management of indeterminate thyroid nodules.
RkJQdWJsaXNoZXIy MTk4NDMw